A composite ESG score offers shaky grounds for a portfolio tilt when six leading versions agree at an average correlation of 0.54. Kaur's review reaches much the same conclusion. The evidence supports a narrower claim than her title's "paradox" framing: specific risks with financial consequences are priced. Composite scores measure those risks noisily.

The trade behind the score

The implied trade is straightforward: rank the cross-section on vendor ESG ratings and overweight the leaders. Its rationale draws on three channels through which sustainability could tell investors something about cash flows or discount rates.

Environmental risks supply the first. Chava finds that banks charge higher rates to firms with environmental concerns; Zerbib estimates that certified green bonds trade at a yield discount of roughly two basis points. Human capital supplies another. Edmans' portfolios of "Best Companies to Work for" firms earned abnormal returns over 25 years, suggesting that markets underprice employee satisfaction. Downside protection completes the case: high-CSR firms beat low-CSR firms during the 2008-2009 crisis (Lins et al.).

Kaur neither built a model nor ran data. Her paper is a structured narrative review, "informed by PRISMA guidelines," using Scopus, Web of Science and SSRN. Publications span 2001 to 2024, and the reference list contains 71 entries. The headline result belongs to Friede et al.: about 90 percent of roughly 2,000 studies find a non-negative ESG-performance relation. Portfolio return studies, however, look much weaker than studies of firm-level operating metrics. Kaur acknowledges this in the abstract: effect sizes "remain modest" and portfolio returns are "weaker, regime-dependent." She attributes the difference to three moderators: industry materiality, firm size and regulatory regime.

We could not test these claims. We lack both the point-in-time histories from several ESG providers and the underlying emissions, safety and employee data, even for US equities. Substituting financial ratios or news sentiment would put an ESG label on a different hypothesis.

Does a healthier company make a better stock?

The financing evidence works against the long side. When green issuers borrow at two basis points less, their bondholders accept two basis points less. The firm's cheaper capital means a lower expected return for its holders. Kaur's carbon evidence makes the corresponding case for brown assets. Bolton and Kacperczyk (2021, 2023) find that high-emission equities offer higher expected returns as compensation for transition risk. The premium remains contested, as Kaur notes in citing Aswani et al. (2024).

Green assets' strong returns in the 2010s therefore need another explanation. Kaur supplies the right one through Pástor et al. (2022): an unexpected increase in climate concern drove that decade's outperformance. Realized returns rose without implying higher expected returns ahead.

Her later discussion of investor motives gets the sign wrong. She presents Pástor et al. (2021) as a model in which investors enjoy owning green assets and accept lower expected returns, "pricing those assets lower in equilibrium." Lower expected returns mean higher prices. Those higher prices deliver precisely the cheap financing she treats as the best-supported environmental finding. That sign matters when sizing a green tilt.

Six ESG raters, different questions

Berg et al.'s 0.54 is the only finding quantified in Kaur's abstract. Across six providers, the average correlation is 0.54, compared with about 0.99 for major credit ratings. Scope and measurement account for roughly 94 percent of the divergence.

The distinction matters. Measurement divergence concerns how raters measure attributes, and better data could reduce that error. Scope divergence concerns which attributes they choose to count. Part of the 0.54 therefore reflects vendors asking different questions about the same company.

More data will not resolve that difference. Kaur cites Christensen et al. (2022), who find that greater corporate disclosure widens disagreement among raters. That disagreement also coincides with higher return volatility. The label has problems of its own: Raghunandan and Rajgopal (2022) find that highly rated ESG funds frequently own firms with worse labor and environmental compliance records than non-ESG funds.

Portfolio results without magnitudes

The ~90 percent figure pools associations across a heterogeneous literature. It gets us little closer to a runnable book. Kaur's portfolio discussion omits the numbers needed to size the trade. Statman and Glushkov's sin-stock exclusion cost is described as "nearly equal" to the tilt's gain, without a figure for either. Brammer et al.'s UK finding, that top-rated stocks earned lower returns, likewise arrives without a magnitude. We did not find a defined screen, rebalance frequency, turnover estimate or cost assumption anywhere in the review.

Positive findings receive the same treatment, including Edmans' 25-year Best Companies portfolio. Edmans over 25 years is "sizable," Eccles et al. over eighteen years "significantly outperformed," and Lins et al. "markedly higher." Even the review's clearest portfolio-level win has no reported magnitude.

There is nothing to benchmark a backtest against.

Kaur acknowledges three limits. She supplies no PRISMA flow count. Her restriction to English-language top-tier journals inherits their bias toward significant results, and she performs no meta-analytic re-estimation. A fourth tension appears in her discussion of gaps. She says "the vast majority" of studies use cross-sectional OLS plagued by endogeneity. On her own description, then, the identified corpus represents a minority of the field.

She calls that corpus "selective rather than exhaustive" and defends the choice: selection "prioritizes causal identification," and "the consistent patterns across studies lend reasonable credibility to the conclusions." Internal consistency in a selective corpus can coexist with poor representation of a field she describes as mostly endogenous OLS.

What would earn a tilt?

The review makes a case for underwriting specific risks. Kaur's recommendation to "triangulate specific underlying indicators" points in that direction too. At 0.54, testing one vendor's point-in-time ratings tests that vendor's objectives. Industry materiality determines which indicators matter: Khan et al. (2016) find that performance on material issues pays, while performance on immaterial issues does not.

Any return must also survive a size control, given Drempetic et al.'s finding that large caps receive higher scores. It must survive the sin-stock exclusion as well, which Hong and Kacperczyk link to higher expected returns.

A composite-score tilt that survives trading costs, a size control and a second rater would move me.